Skip to content

Avoid tmpfs inode exhaustion in scratch worktrees #231

Closed as not planned
Closed as not planned

Description

@shuv1337

Problem

Agent scratchpad worktrees under /tmp use a small tmpfs inode pool. A normal pnpm install can exhaust that pool before the dependency tree is complete, even when byte capacity remains.

Impact

Fresh JJ workspaces fail partway through setup and require a retry on a different location, wasting time and leaving an incomplete environment.

Reproduction

Run pnpm install in a scratchpad JJ workspace under /tmp with a large monorepo dependency tree. The install fails with ENOSPC after roughly 1.6k of 2k packages as the inode limit is reached.

Acceptance criteria

  • Scratchpad guidance defaults dependency-bearing worktrees to a disk-backed path (for example ~/.cache/agent-ws) or performs a clear inode-capacity preflight before installation.
  • The documented path prevents an inode-exhaustion retry for the existing shuv2code workspace setup.
  • Add a focused guard or verification that reports the actionable path when the tmpfs limit is insufficient.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions